godaddy nested data

library(readr)
census_starter <- read_csv("godaddy-microbusiness-density-forecasting/census_starter.csv")
train <- read_csv("godaddy-microbusiness-density-forecasting/train.csv")
test <- read_csv("godaddy-microbusiness-density-forecasting/test.csv")

IDS <- unique(train$cfips)

x_train <- array(0:(39-1), dim=c(39,1)) #Use the first 38 months to train
x_test <- array(38:47, dim=c(10,1)) #Use following 10 months to test

preds <- matrix(0, nrow = length(IDS), ncol=8) #Create matrix of 0's for predictions
last_preds <- matrix(0, nrow = length(IDS), ncol = 8) #Create matrix of 0's for last predictions
lin_trend = 0

ct = 0

for (i,c) in seq_along(IDS) {
  df <- train[train$cfips==c,]
  last <- df$microbusiness_density[length(df$microbusiness_density)]
  active <- 
}

Nested Data

Nested Data Frames by County

library(modelr)
library(tidyverse)
library(dplyr)
library(gapminder)

#Nest

by_county <- train %>%
  group_by(cfips, state) %>%
  nest()

#Model-fitting function

county_model <- function(df) {
  lm(microbusiness_density ~ first_day_of_month, data = df)
}

#Apply to each element

models <- map(by_county$data, county_model)

#Adding models to dataframe

by_county <- by_county %>%
  mutate(model = map(data, county_model))

by_county %>% 
  filter(state == "Texas")

#Adding residuals of each model

by_county <- by_county %>%
  mutate(
    resids = map2(data, model, add_residuals)
  )

#Unnesting

resids <- unnest(by_county, resids)

#Plotting the residuals

resids %>%
  ggplot(aes(first_day_of_month, resid)) +
  geom_line(aes(group = cfips), alpha = 1/3) +
  geom_smooth(se = FALSE)

#Facetting by state

resids %>% 
  ggplot(aes(first_day_of_month, resid, group = cfips)) +
  geom_line(alpha = 1/3) +
  facet_wrap(~state)

#Extracting model quality

glance <- by_county %>% 
  mutate(glance = map(model, broom::glance)) %>%
  unnest(glance)

#Arrange by R square

glance %>% 
  arrange(r.squared)

#Plot worst r squared

glance %>% 
  ggplot(aes(state, r.squared)) +
  geom_jitter(width = 0.5) +
  theme(axis.text.x = element_text(angle = 90))

#Fable #Load packages

library(fable)
library(tsibble)
library(tsibbledata)
library(lubridate)
library(dplyr)

#Keeping tibble format

texas_001 <- by_county %>%
  filter(
    state %in% c("Texas"),
    cfips == "48001"
  )

by_county %>%
  filter(
    state %in% c("Texas"),
    cfips == "48001"
  ) %>%
  model(
    ets = ETS(box_cox(microbusiness_density, 0.3)),
    arima = ARIMA(log(microbusiness_density)),
    snaive = SNAIVE(microbusiness_density)
  ) %>%
  forecast(h ="2 months") %>%
  autoplot(filter(by_county))

t